Earlier this year I wrote about text embeddings / language models and some uses in oil & gas and geoscience Article here
Google is deploying its Bidirectional Encoder Representations from Transformers (BERT) algorithm – it has gone mainstream. It is reported this will improve 1 in 10 searches Google Blog Post here
Language models and dense vectors are unlikely to solve all our search relevancy challenges in the enterprise. However, the potential in very specific areas around Q&A, chatbots and analogue cognitive assistants is potentially very significant.